Skip to contents

Determine Possible Outbreak Detection Methods Based on Available Historic Data

get_possible_methods.Rd

This function identifies which algorithms can be applied for outbreak detection depending on the amount of historic data available for model fitting. The decision is based on the minimum and maximum dates of the time series and the number of weeks reserved at the end of the series (e.g. for current detection).

Usage

get_possible_methods(min_date, max_date, number_of_weeks = 6)

Arguments

min_date

A Date object, the minimum date in the time series used for fitting a model.

max_date

A Date object, the maximum date in the time series used for fitting a model.

number_of_weeks

Integer. Number of weeks at the end of the time series that are reserved for detection / monitoring and therefore not counted as historic data for model fitting. Default is 6.

Value

A character vector of possible algorithm names. Returns NULL if no method is applicable.

Details

Historic data is defined as the period from min_date to max_date - number_of_weeks. The number of available weeks is computed on this interval, counting partial weeks as full weeks to match the weekly aggregation used by the algorithms.

The method selection criteria are approximately:

  • If 4 years (about 208 weeks) or more of historic data are available: all methods are possible.

  • If 3 to <4 years (about 156 to <208 weeks) of historic data are available: all methods except "glm farrington" and "glm farrington with timetrend" are possible.

  • If 2 to <3 years (about 104 to <156 weeks) of historic data are available: all methods except "glm farrington", "glm farrington with timetrend" and "glm harmonic with timetrend" are possible.

  • If 26 to <104 weeks of historic data are available: "Mean", "CUSUM" and "EARS" are possible.

  • If 7 to <26 weeks of historic data are available: "Mean" and "CUSUM" are possible.

  • If 1 to <7 weeks of historic data are available: "CUSUM" is possible.

  • If no training data is available (less than 1 week), NULL is returned.

The function computes max_date_fit = max_date - number_of_weeks and then calculates the number of weeks between min_date and max_date_fit. Partial weeks are counted as full weeks to align with the weekly aggregation of the data. Based on this number of historic weeks, suitable algorithms are selected using available_algorithms.

See also

available_algorithms, get_min_max_date

Examples

if (FALSE) { # \dontrun{
data <- data.frame(
  date_report = seq.Date(Sys.Date() - 500, Sys.Date(), by = "day")
)

mm <- get_min_max_date(data, date_var = "date_report")

get_possible_methods(
  min_date = mm$min_date,
  max_date = mm$max_date,
  number_of_weeks = 6
)
} # }